reduce number of coefficient for polynomial selection with msieve on GPU

[aein]

Hi
I want to do polynomial selection on gpu but the number of coefficient that msieve want do for polynomial selection in very very big.

I know that we can assign each GPU a different range of a1 coefficients using "x,y" in the msieve invocation. if I search in less coefficient (e.g if msieve want do it for 10M coeff but I search in only 7M coeff) how much poly i loss?

if msieve suggested X million coeff, What percentage I can reduce number of coeff that i didn't lost poly too much?!

[VBCurtis]

As far as I know, msieve does not suggest a coefficient range to search; it suggests a number of hours to spend. I don't think it matters what range you search- you should spend "enough" time on poly select, where "enough" is 2-5% of the expected total project length. If you spend 2%, you risk choosing a poly that takes 4% longer to run than one you could have found by spending more time. If you spend 5%, you risk wasting some time because the best poly you found was found halfway through your search.
You can't know ahead of time which one you'll be, but the "best msieve poly scores" thread can give you an idea of the best score you can hope for, given your project's size; if you get within 2-3% of that score, it's unlikely you'll save much time in sieving by finding a better poly.
If the project is a new personal biggest, I try to set a record to post in that thread- but I like playing with the parameters in poly select.

[aein]

when i run polynomial selection i got this in msieve.log:
"searching leading coefficients from 1 to 46231681"

searching in 46231681 coeff in too hard for me in my Gpu.

how can i measure poly that is good enough or not?
what is expected total project length? where is expected total project length?

[VBCurtis]

Originally Posted by aein

when i run polynomial selection i got this in msieve.log:
"searching leading coefficients from 1 to 46231681"

searching in 46231681 coeff in too hard for me in my Gpu.

how can i measure poly that is good enough or not?
what is expected total project length? where is expected total project length?

Project length is what you calculate it to be, based on previous factorizations you've done and the rule of thumb that project length doubles every 5 digits. You should do a couple factorizations in 100-115 digit range to see how things work before you take on something that will occupy your machine for weeks.

A "good enough" poly is also known by experience, though msieve prints into the log an expected range of poly scores (reasonably accurate for most sizes, though it's a bit optimistic for inputs around 150 digits).